During January 2025, Xingjun worked on the ml-explore/mlx-lm repository, focusing on enhancing model management and sourcing flexibility. He developed a feature that enables model snapshot downloading from ModelScope, introducing environment variable toggles to seamlessly switch between ModelScope and Hugging Face Hub. This approach allows teams to adapt model sourcing strategies across different deployment environments, reducing vendor lock-in and streamlining pipeline integration. Xingjun implemented the solution using Python, leveraging his skills in API integration and backend development. The work addressed deployment readiness and operational flexibility, though it was limited in scope to a single feature and did not involve bug fixes.